United States Tornado Analysis

Introduction

• Tornadoes are usually recognizable by their funnel clouds. Typically, this dynamic, funnel-shaped cloud moves beneath the main storm system. Tornadoes have a condensation funnel made up of water droplets, dust, dirt, and debris that makes them visible almost all of the time. Tornadoes are known by a variety of names, including "whirlwind," "windstorm," "cyclone," "twister," and "typhoon," but they are the most dangerous atmospheric storm.

• The dataset provided was derived from a dataset produced by NOAA's Storm Prediction Center and can be found here

We'll explore the following:

• Fatilies
• Most affected states
• Tornadoes by month
• Tornadoes through the years

Let's begin the analysis

1. Data Exploration

#Import the necessary libraries
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import calendar
import plotly.express as px

#Read the dataframe
df = pd.read_csv("us_tornado_dataset_1950_2021.csv")
df.head()

	yr	mo	dy	date	st	mag	inj	fat	slat	slon	elat	elon	len	wid
0	1950	1	3	1950-01-03	IL	3	3	0	39.10	-89.30	39.12	-89.23	3.6	130
1	1950	1	3	1950-01-03	MO	3	3	0	38.77	-90.22	38.83	-90.03	9.5	150
2	1950	1	3	1950-01-03	OH	1	1	0	40.88	-84.58	0.00	0.00	0.1	10
3	1950	1	13	1950-01-13	AR	3	1	1	34.40	-94.37	0.00	0.00	0.6	17
4	1950	1	25	1950-01-25	IL	2	0	0	41.17	-87.33	0.00	0.00	0.1	100

#Renaming the columns for our convenience
df.rename(columns = {'fat':'fatalities', 'len':'length', 'st':'state','inj':'injuries'}, inplace = True)
df.head()

	yr	mo	dy	date	state	mag	injuries	fatalities	slat	slon	elat	elon	length	wid
0	1950	1	3	1950-01-03	IL	3	3	0	39.10	-89.30	39.12	-89.23	3.6	130
1	1950	1	3	1950-01-03	MO	3	3	0	38.77	-90.22	38.83	-90.03	9.5	150
2	1950	1	3	1950-01-03	OH	1	1	0	40.88	-84.58	0.00	0.00	0.1	10
3	1950	1	13	1950-01-13	AR	3	1	1	34.40	-94.37	0.00	0.00	0.6	17
4	1950	1	25	1950-01-25	IL	2	0	0	41.17	-87.33	0.00	0.00	0.1	100

df.shape

(67558, 14)

We can see that there are no null values

df.isna().sum()

yr            0
mo            0
dy            0
date          0
state         0
mag           0
injuries      0
fatalities    0
slat          0
slon          0
elat          0
elon          0
length        0
wid           0
dtype: int64

df.dtypes

yr              int64
mo              int64
dy              int64
date           object
state          object
mag             int64
injuries        int64
fatalities      int64
slat          float64
slon          float64
elat          float64
elon          float64
length        float64
wid             int64
dtype: object

#Checking the unique values for mag columns.The values of -9 indicates the unknown values of tornadoes EF rating
df['mag'].unique()

array([ 3,  1,  2,  4,  0,  5, -9], dtype=int64)

df['mag'].value_counts()

 0    31375
 1    22885
 2     9517
 3     2536
-9      605
 4      581
 5       59
Name: mag, dtype: int64

Here we plot the number of tornadoes each year since 1950.

#Grouping by year
df1 = df.groupby('yr')['yr'].count()
plt.plot(df1)
plt.xlabel('Year')
plt.ylabel('Number of Tornadoes')
plt.title('Tornadoes in the US',fontweight='bold')
#plt.text(1, 1, 'testing captions', ha='center')

Text(0.5, 1.0, 'Tornadoes in the US')

Obervations:

• The number of tornadoes have consistently increased over the since 1950, although there have been dips in-between

Plotting the top 10 states with the highest number of tornadoes

df2 = df.groupby('state')['state'].count().reset_index(name = 'counts').sort_values(['counts'],ascending = False)
df2.head(10).plot(x = 'state', kind = 'bar')
plt.xlabel('State')
plt.ylabel('Number of tornadoes')
plt.title('Top 10 states with the highest number of tornadoes',fontweight='bold')

Text(0.5, 1.0, 'Top 10 states with the highest number of tornadoes')

Observations:

• We can observe from the above graph that the southern, south-eastern and mid-western regions have seen the highest number of tornadoes

Here's an interactive map of the US with the counts of tornadoes in each state.

fig = px.choropleth(df2,
                     locations = 'state',
                     locationmode = 'USA-states',
                     color = 'counts',
                     color_continuous_scale = 'REDS',
                     scope = 'usa',
                    
                   )

fig.update_layout(title_text='Count of tornadoes across the states in the US.<br>(Hover for statewise counts)')

fig.show()

Plot for monthly tornadoes in the US

df_months_count = df.groupby('mo')['mo'].count().reset_index(name = 'monthly_tornadoes')
df_months_count['mo'] = df_months_count['mo'].apply(lambda x: calendar.month_abbr[x])
df_months_count.plot(x = 'mo', kind = 'bar')
plt.xlabel('Months')
plt.ylabel('Number of tornadoes')
plt.title('Number of tornadoes per month throughout the US',fontweight='bold')

Text(0.5, 1.0, 'Number of tornadoes per month throughout the US')

Obervation:

• The months of May and June have seen the highest numbers of tornadoes

Fatalities

print('The total number of fatalities as a result of tornadoes:',df.fatalities.sum())

The total number of fatalities as a result of tornadoes: 6112

Plot for Top States with highest fatalities

fatalities = df.groupby('state')['fatalities'].sum().reset_index().sort_values('fatalities', ascending = False)
fatalities.head(10).plot(x = 'state', kind = 'bar')
plt.xlabel('State')
plt.ylabel('Fatalities')
plt.title('Top 10 states with the highest fatalities',fontweight='bold')

Text(0.5, 1.0, 'Top 10 states with the highest fatalities')

Obervations:

• The states with highest number of tornadoes have seen the most fatalities

Here's the plot for average fatalities per EF rating

ef_ratings = df.iloc[np.where(~df.mag.isin([-9]))]
ef_ratings = ef_ratings.groupby('mag')['fatalities'].mean().reset_index()
ax = ef_ratings.plot(x = 'mag',y ='fatalities',kind = 'bar')
ax.set_xlabel("F or EF Rating")
ax.set_ylabel("Mean fatailies")
ax.set_title('Mean fatalities for each EF rating',fontweight='bold')

Text(0.5, 1.0, 'Mean fatalities for each EF rating')

Observations:

What is an EF scale?\ The Enhanced Fujita scale (abbreviated as EF-Scale) rates tornado intensity based on the severity of the damage they cause.

• We can see that the tornadoes with an EF ratings of more than 2 have caused fatilites.

Plot for the fatalities throughout the US since 1950

fats_per_yr  = df.groupby(['state','yr'])['fatalities'].sum().reset_index()
fig = px.choropleth(fats_per_yr, 
                    locations = "state", 
                    
                   locationmode="USA-states", 
                   scope="usa",
                   color = 'fatalities' ,
                   
                   labels={'yr':'Year','state':'State','fatalities':'Deaths'},
                   color_continuous_scale= 'YlOrBr',animation_frame = "yr")
                   
fig.update_layout(title = "Tornado Fatalities in States Per Year")
fig.show()

Plot for fatalities over the years in the US

df_fat = df.groupby('yr')['fatalities'].count().reset_index()
df_fat.plot('yr', 'fatalities')
plt.xlabel('Year')
plt.ylabel('No of fatalities')
plt.title('Fatalities over the years',fontweight = 'bold')

Text(0.5, 1.0, 'Fatalities over the years')

Observation:

• The number of fatalities have overall increased progressively over the years

state_fatalities = df.groupby('state')['fatalities'].sum().reset_index()
fig = px.choropleth(state_fatalities,
                     locations = 'state',
                     locationmode = 'USA-states',
                     color = 'fatalities',
                     color_continuous_scale = 'ORRD',
                     scope = 'usa',
                    
                   )

fig.update_layout(title_text='Count of tornadoes across the states in the US.<br>(Hover for statewise counts)')

fig.show()

Observations:

• The southern and midwestern states have seen more fatalities
• 8 states have no fatalities

CONCLUSION:

• There have been 67558 tornadoes in total.
• The southern and the midwestern US is most affected by the tornadoes.
• Tornadoes with the rating of EF5 is responsible for most fatalities.
• May saw the most tornadoes touch down by month, followed by June.